EUCA dataset

EUCA 数据集描述 相关论文：EUCA：以最终用户为中心的可解释 AI 框架作者：Weina Jin、Jianyu Fan、Diane Gromala、Philippe Pasquier、Ghassan Hamarneh 简介：EUCA 数据集用于建模个性化或交互式可解释 AI。它包含 32 个最终用户对 12 种解释形式（包括基于特征、示例和基于规则的解释）的偏好的 309 个数据点。这些数据是从对 2019-2020 年大温哥华市区 32 名非专业参与者的用户研究中收集的。在用户研究中，参与者 (P01-P32) 接受了人工智能辅助的关键任务，包括房价预测、健康状况预测、购买自动驾驶汽车和学习生物检查 [1]。在每个任务中并针对其给定的解释目标 [2]，参与者选择并排列他们认为最合适的解释形式 [3]。 1 EUCA_EndUserXAI_ExplanatoryFormRanking.csv 列描述：索引-参与者人数案例-任务-解释目标组合接受使用AI？相信它？ - 鉴于任务和解释目标需要解释，参与者回应他们是否会使用人工智能？ - 参与者回答他们是否要求对AI 1、2、3、...进行解释 - 解释表卡选择和排名卡是否满足要求？ - 卡片选择后，参与者被问及选择的卡片组合是否满足他们的可解释性要求。 2 EUCA_EndUserXAI_demography.csv 它包含参与者的人口统计数据，包括他们的年龄、性别、教育背景以及他们对人工智能的知识和态度。 EUCA 数据集 zip 文件供下载 EUCA 数据集的更多上下文 [1] 关键任务 有四个任务。任务标签及其对应的任务标题分别为：房子 - 卖房车 - 购买自动驾驶汽车 健康 - 个人健康决定鸟 - 学习鸟类物种 中呈现的任务故事板和解释目标请参考 EUCA 量化数据分析报告用户研究。 [2] 解释目标最终用户可能有不同的目标/目的来检查人工智能的解释。 EUCA数据集包括以下11个解释目标，其在数据集中的【标签】、全称和描述【信任】校准信任：信任是建立人-AI决策伙伴关系的关键。由于用户很容易不信任或过度信任人工智能，因此校准信任以反映人工智能系统的能力非常重要。 [安全] 确保安全：用户需要确保安全的决定后果。

EUCA Dataset Description
Related Paper: EUCA: An End-User Centered Explainable AI Framework
Authors: Weina Jin, Jianyu Fan, Diane Gromala, Philippe Pasquier, Ghassan Hamarneh

Introduction: The EUCA dataset is developed for modeling personalized or interactive explainable AI. It contains 309 data points capturing 32 end-users' preferences across 12 explanation formats, including feature-based, example-based, and rule-based explanations. These data were collected from a user study involving 32 non-professional participants in the Greater Vancouver Area between 2019 and 2020. During the study, participants (labeled P01-P32) completed AI-aided critical tasks: housing price prediction, health status prediction, autonomous vehicle purchasing, and bird species learning [1]. For each task and its specified explanation goal [2], participants selected and ranked the most appropriate explanation formats [3].

1. EUCA_EndUserXAI_ExplanatoryFormRanking.csv
Column Descriptions:
- Index: Participant number
- Case: Combination of task and explanation target
- Accept Use of AI?: Whether the participant indicates whether they would use the AI system given the task and explanation goal requiring explanations
- Trust It?: Whether the participant reports whether they request explanations for AI 1, 2, 3, ...
- Does the Selected and Ranked Explanation Card Set Meet Requirements?: After selecting and ranking the explanation cards, participants were asked whether the chosen card combination satisfied their explainability requirements.

2. EUCA_EndUserXAI_demography.csv
This file contains participants' demographic information, including age, gender, educational background, as well as their knowledge of and attitudes towards artificial intelligence.

Downloadable ZIP File of the EUCA Dataset
Additional Context of the EUCA Dataset
[1] Critical Tasks: There are four tasks, with their labels and corresponding titles as follows:
- House: Housing price prediction
- Car: Autonomous vehicle purchasing
- Health: Personal health decision-making
- Bird: Bird species learning
Please refer to the EUCA Quantitative Data Analysis Report for the task storyboards and explanation targets presented in each task.
[2] Explanation Goals: End-users may have diverse goals for inspecting AI explanations. The EUCA dataset includes the following 11 explanation goals, with their labels, full names and descriptions in the dataset:
- [Trust] Calibrate Trust: Trust is critical for establishing human-AI decision-making partnerships. Users may easily distrust or over-trust AI systems, so calibrating trust to reflect the actual capabilities of the AI system is essential.
- [Safety] Ensure Safety: Users need to confirm that the consequences of their decisions are safe.

搜集汇总

背景与挑战

背景概述

EUCA数据集是一个用于个性化可解释AI建模的数据集，包含32名参与者对12种解释形式的偏好数据，共309个数据点，这些数据来源于2019-2020年大温哥华地区的用户研究，涉及房价预测、健康状况预测等四个关键任务和11个解释目标。数据集还提供了参与者的人口统计信息，如年龄、性别和教育背景。